I was speaking to someone this week about our segmentation or scoring process – and it’s different than most.
What is scoring? It's where you assign a value or score to an overdue account -- that score could be based on a credit bureau score, it could be based on payment patterns, really, it's a way to rank and prioritize accounts that are more likely to pay if contacted.
In the past we used a credit bureau for segmentation scoring, and it did improve our performance by about 10%, but it still had a lot of oddities, and was a single point in time snapshot, not an evolving score. So over a few years we built our own predictive model, ran it alongside the bureau score, and built something better that replaced it. And our segmentation formula is constantly evolving as we take in new data, and reverse engineer from payments where the money is coming from.
Why would we score our collection files? Well, if you have a client that liquidates 60% and only gives you a couple hundred accounts a month, you might easily be able to create a positive ROI on human effort, and scoring might not be necessary – but what if the liquidation is 5%? What if a client drops 30,000 accounts on you every month? Do you want your collection team spending 95% of their time and effort on files that won’t pay? Do you have the manpower to effectively put eyeballs on all 30,000 accounts? If you score files, you can create different strategies for high, medium and low scoring files.
Add to that, if your score is a moving target, you can decide when a file should stop having intensive human effort, or if new data comes in that raises the score, elevate that file immediately to a team member’s attention.
What’s The Formula?
So here's what we use.
If you have a hypothetical client that historically liquidates 20%, all other things being equal, that means the law of averages means every file has a 20% chance of recovery. This is that client’s baseline score (we multiply it by 10, so a base score of 200).
Everything beyond that baseline is a variance up or down – young people, files with multiple forms of contact such as phone numbers or email, accounts with more recent dates of the debt incurred, recent payments, mortgage holders, consumers without 27 terrible R9 tradelines on their bureau, all of those are positive scores. Very aged accounts, older consumers, consumers with other collection items or judgments on their bureau would reduce the score.
And that score can be a moving target based on behaviours or patterns – if you call someone about a collection file, and they tell you to take a ride in a handbasket to heck, that probability score should drop immediately. If calls or SMS message or emails go unanswered, that score should depreciate for every unreturned call, and if a consumer makes a partial payment or returns calls the score should go up based on positive actions.
And of course, your scoring or segmentation algorithm should always be evolving. The best thing we did when we built our model in parallel with the bureaus is reverse engineer what was working – look at all the accounts that were paying and built a mathematical formula that placed values on home ownership, responses, consumer age, etc. It’s not perfect, and you should always look at the low scoring files that are paying and ask ‘why?’
How Scoring Accounts Works In Practice
Collections is a manpower intensive process, no matter how many predictive dialers or virtual AI agents you add – and 80% of human effort typically does not end up in collecting funds. Anything that can improve those odds, even a little, significantly improves the revenue to manpower ratio. Improving your recoveries by 2% of your human effort works out to a 10% increase in revenues, if you follow my 80%/20% math.
If you can move the needle on effectiveness, this can make your company more profitable, which means they can reinvest in tools, infrastructure, staff retention, and all that. Yes, if you are a collection agency you can improve a client’s recoveries, and make the recovery curve for that client move more quickly, but by being able to focus manpower in the right direction means your team members are spending their time and expertise more meaningfully.
We live in a world now where convincing someone to pay an account is not the biggest hurdle, it’s getting them to communicate, answer the phone, or respond to an email. Scoring lets you focus on accounts that are likely more responsive, or have proven themselves to be more responsive.
This doesn’t mean not working lower scoring accounts, it means using a different workflow – mass texting or emails on low probability accounts until they get an initial response, and then having a human intervening …
The other advantage to segmentation or scoring is you can build a model in advance of the work – if a new client comes on, you can score the accounts initially through all your known initial metrics if you already service clients in that same industry, and get a good sense of the aggregate quality of the paper on multiple values, and come up with an initial work flow, even if you don’t have a historical baseline liquidation for that specific client.
Creditors and agencies that historically need to service large volumes of accounts can set up scoring to approach their accounts strategically, where they aren’t just working highest to lowest balances, they look at probabilities. That means getting a better handle on large volumes and finding the accounts willing to pay faster.
For a sense of scale and my personal viewpoint, since January 1st to today, in about three months our company has processed 49,266 payments. Of those payments, 33,106 had scored highly (what we predict has a 40%+ probability of paying on contact) and had human intervention. Of the remaining payments, 7,188 self-cured through our website or a payment link, and the remaining 8,972 were low scoring files that paid through human intervention. That shows we spent our time wisely.
I did not build our scoring algorithm in a vacuum -- I've had lots of great conversations with folks in our industry and shared ideas. If you want to chat about segmentation, always happy to talk ... drop me an email or give me a call.
Thanks kindly,
Blair DeMarco-Wettlaufer
KINGSTON Data & Credit
226-946-1730
blair@receivableaccounts.com
